Cache handles stale values

[bargulg]

This is needed for any reasonable cache that can set expire time on keys and remove them automatically. For example redis

[wtfrank]

Isn't it a bit weird to put the expiry time as an offset from the current time, rather than an absolute time: e.g. "ex + time.time()"? Cos small processing delays will push the calculated expiry time slightly into the future, which wouldn't happen with an absolute time.

[bargulg]

well, from the API we get cache time as an offset, and i don't see any real reason to translate it, since it would make things more complicated and those processing delays are very small... it would be different if cache time was shorter, but for something like 5 minutes, it doesn't matter

[jonobrien]

has anyone tested this pr?

[wtfrank]

Do you think it would be useful to expose the expiry time (and the original object creation/cache time) to the user somehow? For example, if I want to track a character's location in a wormhole mapping application then I would want to fire off a new location request as soon as the previous one had expired.

[jonobrien]

That use-case sounds useful definitely, might need more testing to see what is possible.

[bargulg]

i've been using my own version with this patch for half a year now, and it works and it makes it a lot easier to create real caching mechanisms... but now there's been some change adding memcached to the code of of pycrest so that would need to be adapted to this patch, too... and i don't care enough about pycrest to fix it anymore, because this has been waiting here for half a year

[hkraal]

@bargulg it's a shame it hasn't been picked up back in the day and your efforts went to waste due changes along the way.

That being said a feature like this will find it's way into the next pycrest release if it's up to me. My concept ( #37 ) results in an dictionary being stored as a whole in the cache. In that dictionary there will be an "expire_in" and "expire_at" property on which the caching can take his course of action using the relative or absolute time.
It's intended to fix a few other things as well like providing CREST endpoint version information which seems to be a popular request.

I've opened #38 to track this specific feature so we can, if you agree, close this PR for now.